Docs / v1.0 · hgt-v1 · ibm it-aml medium

Reading the topology
of laundered money
one edge at a time.

A heterogeneous graph transformer for wire-transfer AML, trained on IBM's IT-AML Medium NeurIPS 2023 dataset. Inline attention explanations, sub-200 ms CPU inference, and a D3 visual surface — documented end to end.

LIVE  ·  subgraph inference
k = 2 edges 247 flagged 3
x+ attention hetero · 3 node types
Edge e_1873
Illicit p 0.842
Tier HIGH
Latency p95 147 ms
Model hgt-v1
individual business bank
wire attention flagged